130 likes | 254 Views
Planning Motions with Intentions. A presentation on the paper Planning Motions with Intentions written by Yoshihito Koga, Koichi Kondo, James Kuffner and Jean-Claude Latombe. By Chris Montgomery. What is a “Motion with Intention?”. A motion for accomplishing some task Non-predictable
E N D
Planning Motions with Intentions A presentation on the paper Planning Motions with Intentions written by Yoshihito Koga, Koichi Kondo, James Kuffner and Jean-Claude Latombe By Chris Montgomery
What is a “Motion with Intention?” • A motion for accomplishing some task • Non-predictable • Cannot be described by laws physics • Solves a problem
Our Goal • To Realize the automatic generation of human and robot arm motions to complete manipulation tasks
Our Problem • Multi-Arm Manipulation Problem • Must find a collision free path for the arms to grasp and then carry some specified moveable object from its initial location to its desired goal • Must account for the ability of the arms to change their grasp on the object • Necessary to solve all problems
Necessary Information to Create a New Path Planner • Geometry of the Environment • Initial and Goal Configurations • Set of Potential Grasps • Inverse Kinematics of the Arms
Definitions • Stable Space • The set of all legal configurations where the object is statically stable between contacts (arms, obstacles, etc) • Grasp Space • The set of all the configurations where one or more of the arms rigidly grasp the object in a way that there is sufficient torque to move the object
More Definitions • A Transit Path • Movement of an arm that does not move the object. • A Transfer Path • Movement of an arm(s) that moves the object. Some non-grasping arms may also be moving to avoid collisions.
More Definitions • Manipulation Path • An alternate sequence of transit and transfer paths that connect an initial system configuration to a goal system configuration
Two-Stage Planning Approach • Generate a series of subtasks to achieve the system goal configuration • Plan a transit or transfer path for each subtask • But how can one determine whether a subtask can be completed without actually completing it?
Generating a Path • While path is planned, arms are checked to ensure the object can be grasped • At every point along the path, the planner keeps a list of possible grasps
Generating a path • Once a collision free path is found, the planner determines which grasp is valid longest • Transit path is planned with this grasp • If grasp will not last until the final configuration, right before the grasp becomes invalid the planner is used again
Limitations • Transit task between two transfer tasks is difficult to solve when ungrasp/regrasp is necessary • Intermediate step is necessary
Limitations to this process • Does not plan for regrasps at configurations where the object makes contact with an obstacle • Only probabilistically complete • Computation time cannot be bounded in advance—could run forever • However, it usually returns a path quickly, so an arbitrary time can be set to determine when a path is not possible